OpenAI released a software update for its Codex and Work tools that allows developers to adjust reasoning levels mid-conversation without clearing the prompt cache. The change enables users to switch to lower reasoning for simple tasks and increase it for complex ones without reprocessing the conversation history, reducing the need to rebuild context.
The company also updated the Codex CLI to version 0.164.0, introducing reusable prompt presets that save model and reasoning settings for recall via /presets or Ctrl X P. The CLI release includes inline model and permission changes, fixes for long session resumes, and support for Windows proxy authentication and Enterprise MCP sign-in. Additionally, connection speeds have been increased for persistent agents known as dots.